clear cached templates between specs #864

JonRowe · 2013-12-06T00:00:09Z

One of our cukes was failing because Rails caches templates and the differing
format seems not to be enough to invalidate it automatically, I'm going to
open a Rails patch for this but in the mean time this fixes that cuke.

myronmarston · 2013-12-06T00:22:18Z

lib/rspec/rails/example/view_example_group.rb

@@ -152,6 +152,9 @@ def _include_controller_helpers
          view.lookup_context.prefixes << _controller_path
        end

+        # fixes bug with differing formats
+        view.lookup_context.view_paths.map(&:clear_cache)


This could be each(&:clear_cache), which I think makes more sense since you aren't storing the array returning by map.

No idea why I used map, I'll change it when the build finishes... (I want to see the state of it after)

Woop! Green, I'll change this now then, wait for a second green then merge :)

One of our cukes was failing because Rails caches templates and the differing format seems not to be enough to invalidate it automatically, I'm going to open a Rails patch for this but in the mean time this fixes that cuke.

clear cached templates between specs

clear cached templates between specs Fixes the 2-99 build, see #864

The caching was turned off in [1] because different formats were being cached under the same key. I believe the need to do this was removed in the previous commit. This reduces the time it takes to run FutureLearn’s view specs by about 50%. I think it’s sensible to use the view cache during test because it’s normally on between requests in other environments, and it looks like it was only turned off to work around this caching issue. [1] rspec#864

Rails expects the format to be a symbol. It filters formats against `Mime::SET.symbols` so formats which are unknown (or are known but are of an unexpected type) are not used in a cache keys for the Rails' `ActionView::Resolver::Cache`. This made the view cache unusable when there was multiple templates with the same name but different format and was the the reason the cache was disabled in [1]. [1] rspec#864

The caching was turned off in [1] because different formats were being cached under the same key. I believe the need to do this was removed in the previous commit. This reduces the time it takes to run FutureLearn’s view specs by 50%. I think it’s sensible to use the view resolver cache during tests because it’s normally on between requests in other environments including when creating tests with `ActionView::TestCase` in a new Rails application. It looks like it was only turned off in rspec-rails to work around this caching issue. [1] rspec#864

Rails expects the format to be a symbol. It filters formats against `Mime::SET.symbols` so formats which are unknown (or are known but are of an unexpected type) are not used in a cache keys for the Rails' `ActionView::Resolver::Cache`. This made the view cache unusable when there was multiple templates with the same name but different format and was the the reason the cache was disabled in [1]. This isn't the only way of fixing this problem- for example Rails could be changed to ensure symbols aren't compared with strings as was suggested in [2], but I feel that it's rspec-rails' responsibility to work with the existing Rails API. [1] rspec#864 [2] rails/rails#13461

The caching was turned off in [1] because different formats were being cached under the same key. I believe the need to do this was removed in the previous commit. This reduces the time it takes to run FutureLearn’s view specs by 50%. I think it’s sensible to use the view resolver cache during tests because it’s normally on between requests in other environments including when creating tests with `ActionView::TestCase` in a new Rails application. It looks like it was only turned off in rspec-rails to work around this caching issue. [1] rspec#864

Rails expects value of the `:format` option be a symbol. It filters formats against `Mime::SET.symbols` so formats which are unknown (or are known but are of an unexpected type) are not used in cache keys for the Rails' `ActionView::Resolver::Cache`. This made the resolver cache unusable when there was multiple templates with the same name but different formats and was the reason the cache was disabled in rspec-rails [1]. This isn't the only way of fixing this problem- for example Rails could be changed to ensure symbols aren't compared with strings as was suggested in [2], but I feel that it's rspec-rails' responsibility to work with the existing Rails API. [1] rspec#864 [2] rails/rails#13461

The caching was turned off in [1] because different formats were being cached under the same key. I believe the need to do this was removed in the previous commit. This reduces the time it takes to run FutureLearn’s view specs by 50%. I think it’s sensible to use the view resolver cache during tests because it’s normally on between requests in other environments including when creating tests with `ActionView::TestCase` in a new Rails application. It looks like it was only turned off in rspec-rails to work around this caching issue. I haven't added any test coverage in this change since it feels like it would be testing a negative (the resolver cache isn't cleared). [1] rspec#864

Rails expects value of the `:format` option be a symbol. It filters formats against `Mime::SET.symbols` so formats which are unknown (or are known but are of an unexpected type) are not used in cache keys for the Rails' `ActionView::Resolver::Cache`. This made the resolver cache unusable when there was multiple templates with the same name but different formats and was the reason the cache was disabled in rspec-rails [1]. This isn't the only way of fixing this problem- for example Rails could be changed to ensure symbols aren't compared with strings as was suggested in [2], but I feel that it's rspec-rails' responsibility to work with the existing Rails API. [1] rspec#864 [2] rails/rails#13461

The caching was turned off in [1] because different formats were being cached under the same key. I believe the need to do this was removed in the previous commit. This reduces the time it takes to run FutureLearn’s view specs by 50%. I think it’s sensible to use the view resolver cache during tests because it’s normally on between requests in other environments including when creating tests with `ActionView::TestCase` in a new Rails application. It looks like it was only turned off in rspec-rails to work around this caching issue. I haven't added any test coverage in this change since it feels like it would be testing a negative (the resolver cache isn't cleared). [1] rspec#864

Rails expects value of the `:format` option be a symbol. It filters formats against `Mime::SET.symbols` so formats which are unknown (or are known but are of an unexpected type) are not used in cache keys for the Rails' `ActionView::Resolver::Cache`. This made the resolver cache unusable when there was multiple templates with the same name but different formats and was the reason the cache was disabled in rspec-rails [1]. This isn't the only way of fixing this problem- for example Rails could be changed to ensure symbols aren't compared with strings as was suggested in [2], but I feel that it's rspec-rails' responsibility to work with the existing Rails API. [1] rspec#864 [2] rails/rails#13461

The caching was turned off in [1] because different formats were being cached under the same key. I believe the need to do this was removed in the previous commit. This reduces the time it takes to run FutureLearn’s view specs by 50%. I think it’s sensible to use the view resolver cache during tests because it’s normally on between requests in other environments including when creating tests with `ActionView::TestCase` in a new Rails application. It looks like it was only turned off in rspec-rails to work around this caching issue. I haven't added any test coverage in this change since it feels like it would be testing a negative (the resolver cache isn't cleared). [1] rspec#864

myronmarston reviewed Dec 6, 2013
View reviewed changes

JonRowe mentioned this pull request Dec 6, 2013

fixing the controller spec generator to not emit deprecation warnings #862

Merged

blow away the view cache between specs

96d3e75

One of our cukes was failing because Rails caches templates and the differing format seems not to be enough to invalidate it automatically, I'm going to open a Rails patch for this but in the mean time this fixes that cuke.

JonRowe mentioned this pull request Dec 6, 2013

Use details to invalidate template resolver cache. rails/rails#13208

Closed

JonRowe added a commit that referenced this pull request Dec 6, 2013

Merge pull request #864 from rspec/fixup_view_caching

e338e8d

clear cached templates between specs

JonRowe merged commit e338e8d into master Dec 6, 2013

JonRowe deleted the fixup_view_caching branch December 6, 2013 02:52

alindeman mentioned this pull request Dec 8, 2013

Set cache_key correctly for Rails 4 #868

Closed

3 tasks

JonRowe added a commit that referenced this pull request Dec 14, 2013

Merge pull request #864 from rspec/fixup_view_caching

13df8bb

clear cached templates between specs Fixes the 2-99 build, see #864

zetter mentioned this pull request Sep 3, 2015

Halve the time it takes to run view specs by enabling the resolver cache #1452

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

clear cached templates between specs #864

clear cached templates between specs #864

Uh oh!

JonRowe commented Dec 6, 2013

Uh oh!

myronmarston Dec 6, 2013

Uh oh!

JonRowe Dec 6, 2013

Uh oh!

JonRowe Dec 6, 2013

Uh oh!

Uh oh!

Uh oh!

clear cached templates between specs #864

clear cached templates between specs #864

Uh oh!

Conversation

JonRowe commented Dec 6, 2013

Uh oh!

myronmarston Dec 6, 2013

Choose a reason for hiding this comment

Uh oh!

JonRowe Dec 6, 2013

Choose a reason for hiding this comment

Uh oh!

JonRowe Dec 6, 2013

Choose a reason for hiding this comment

Uh oh!

Uh oh!